18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
10.1146/annurev.fluid.35.101101.161144
C. P. Caulfield
Department of Mechanical and Aerospace Engineering, University of California,
San Diego, La Jolla, California 92093-0411; e-mail: cpc@mae.ucsd.edu
1. INTRODUCTION
In comparison with the relatively simple problem of thermal convection, the problem of understanding the mechanics of the turbulence transition in an initially laminar free shear layer involves a considerable increase in complexity. Whereas
in the problem of thermal convection the strength of the thermal forcing is most
often kept fixed by fixing the appropriate Rayleigh number and then measuring the
corresponding Nusselt number (nondimensional heat transfer) in a statistical equilibrium state, in the case of the free shear layer, mixing is forced to occur through a
highly time-dependent sequence of specific events that originate through a simple
mechanically driven instability whose strength depends not only on the magnitude
but also the structure of the vertical shear in horizontal velocity. In this problem
the correct nondimensional measure of the strength of the initial mechanical forcing is the so-called gradient Richardson number Ri(z), which is defined as
Ri(z) =
0066-4189/03/0115-0135$14.00
(gd`og(z)/dz)
N2
=
,
2
(dV/dz)
(dV/dz)2
(1)
135
1 Nov 2002
18:38
136
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
is the height
variation of the density field in the initially parallel-flow basic state, and V(z) is
the initial height variation of background horizontal velocity. The important theorem of Miles (1961) and Howard (1961) assures us that this stratified parallel
flow will be stable against small amplitude fluctuations so long as Ri(z) > 0.25
at all heights. However, as soon as Ri(z) < 0.25 at some level in the parallel
flow, a rich variety of instabilities may occur (see Drazin & Reid 1981). Although
circumstances in which the stratified shear instability that arises when the MilesHoward criterion for stability is violated may produce coherent structures other
than the vortical disturbances of Kelvin and Helmholtz (KH) (see Drazin & Reid
1981), the development of these so-called KH instabilities appears to constitute
the usual first step toward the onset of vigorous mixing of the density field, and
they have been observed both in the atmosphere and oceans (see, e.g., Woods
1968, De Silva et al. 1996). KH instabilities are the inevitable consequence of
instability in circumstances in which either the stratification is sufficiently weak
or the thickness of the layer of strong, inflectional shear (see Drazin & Reid
1981) is similar to the thickness of any coincident concentration in the (stable)
density gradient. Only when the latter thickness is significantly smaller than the
former, and the stratification is sufficiently strong, does the characteristic instability bifurcate into the more exotic Holmboe instability (Holmboe 1962, Browand
& Winant 1973, Koop & Browand 1979, Smyth et al. 1988, Lawrence et al.
1991).
The Holmboe instability is characterized at finite amplitude (see Smyth et al.
1988) by vortical disturbances displaced above and below the region of strong
density gradient that propagate in opposite directions relative to the mean flow
unlike the KH instability, which is stationary in a frame of reference moving with
the mean flow. The Holmboe instability also induces significant overturning of
the region of strong density gradient. Experimental evidence (Browand & Winant
1973, Koop & Browand 1979, Strang & Fernando 2001) suggests that flows that
exhibit Holmboe instabilities alone cause relatively weak mixing compared to the
intense mixing associated with high Reynolds number and low Richardson number
KH instabilities. However, some recent numerical evidence (Smyth & Winters
2002) suggests that the long time evolution of the instability (in the absence of
transition) may nevertheless induce significant mixing, and experimental evidence
(see Strang & Fernando 2001) suggests that the Holmboe instability may play a
significant role in enhancing the mixing associated with KH breakdown at higher
Richardson numbers.
Because KH instabilities may induce extremely vigorous mixing through intense three-dimensional turbulence when transition occurs, we focus the discussion here on explaining the mechanisms by which this can occur, concentrating
on recent advances in the explanation and quantification of the time-dependent
mixing behavior obtained through theoretical and numerical analyses. Because
the dynamical processes that occur during the evolution of an unstable free shear
layer feed back in a strongly negative fashion on the height-dependent horizontally
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
137
averaged mean state that supports the initial Ri < 0.25 KH instability, if the initial
mean state is not continuously reinforced the process of shear-layer transition will
have a finite lifetime, the end state of which will be a relaminarized mean state
that is dynamically stable. This is a fundamental point that distinguishes unforced
stratified shear flow evolution from forced stratified turbulence, whether the latter is realized experimentally (see Linden 1979, Fernando 1991 for reviews and,
more recently, Park et al.1994, Keller & Van Atta 2000) or numerically through
the maintenance of the mean shear and density profiles (Gerz et al. 1989, Holt
et al. 1992, Jacobitz et al. 1997, Diamessis & Nomura 2002). As we discuss in
Section 3, several important points of connection do exist between these distinct
circumstances, particularly in flows with sufficiently weak stratification and large
Reynolds number, as in such cases the decay of the turbulence following transition
is slow enough to allow for meaningful comparison.
Analysis of the processes that occur during mixing layer development point
to the utility of consideration of a discrete sequence of interactions, with the
character of the flow passing through an identifiable life cycle. In order to isolate the
fundamental character of these distinct stages, it is most straightforward to assume
that the flow remains horizontally periodic and that the initial instability develops in
time, rather than in space as it does in experiments in which the instability is forced
to develop downstream from a splitter plate (see, e.g., Lawrence et al. 1991). Such
analyses are therefore most directly applicable to understanding the measurements
made in tilted-tube experiments such as those by Thorpe (reviewed in Thorpe
1987). Our specific goal in this review, however, is to discuss the way in which
recent theoretical developments (discussed initially in the papers by Winters et al.
1995 and Winters & DAsaro 1996, which built upon the seminal ideas of Lorenz
1955) are enabling us to make quantitative contact between numerical simulations
of shear flow transition (in particular those by Caulfield & Peltier 2000 and Smyth
& Moum 2000a,b) and experimental observations (Linden 1979, Fernando 1991,
Strang & Fernando 2001). The accumulating data present compelling evidence
that, provided the flow undergoes a true transition to turbulence, the efficiency (i.e.,
the ratio of irreversible increases in the potential energy of the density distribution
to the irreversible loss of kinetic energy) is approximately 20%.
The identification of the sequence of secondary instabilities that control the
evolution of the free shear layer, as well as that for the unstratified end member,
began with the work by Peltier et al. (1978), Davis & Peltier (1979), and Klaassen
& Peltier (1985b, 1989, 1991). In these papers it was first suggested on physical grounds, and then demonstrated mathematically, based on the application of
rigorous Floquet methods, that once the initially two-dimensional stratified KH
instability saturated, it became susceptible to a three-dimensional convective instability that was focused in the regions around the periphery of the vortex cores,
or billows that characterize the cats eye structure of the saturated KH instability
(see Figure 1, where the regions of static instability induced by the primary instability are shaded). The validity of these three-dimensional linear analyses of the
stability of the two-dimensional nonlinear KH billow relied on the assumption of a
1 Nov 2002
18:38
138
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
Figure 1 Contours of spanwise vorticity at a time when the primary KH billow has maximum amplitude in a stratified shear flow with initial hyperbolic tangent velocity and density
profiles. Re = 750, Pr = 1, and minimum gradient Richardson number Ri(0) = 0.05. Statically unstable regions are shaded. [Adapted from Figure 1b in Caulfield & Peltier (2000),
used with permission.]
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
139
for transition in the unstratified case (see Kerswell 2002 for a thorough review of
the elliptical instability).
As we review in detail here, the efficiency of the mixing that occurs immediately
after the initially two-dimensional nonlinear KH instability matures is rather high,
typically of the order of 70%. Crucially, this mixing is associated with convective
overturning of the statically unstable regions in the periphery of the primary billow,
which trigger the onset of the three-dimensional convective instability, and is
efficient because the flow is in no way turbulent at this stage. Therefore, the
viscous dissipation is relatively close to its laminar value, and thus a significant
proportion of the energy being lost irreversibly from the velocity shear is leading
to irreversible increases in flow potential energy. The relative importance of this
stage is, unsurprisingly, strongly dependent on the flow Reynolds number and
overall stratification, as we show below, and for sufficiently small Richardson
numbers, and sufficiently large Reynolds numbers, the mixing associated with
this stage can be completely overwhelmed by turbulent mixing associated with
the breakdown of the finite amplitude saturated structures owing to the convective
instability.
This point, that the dominant mixing processes for a stratified shear flow are
driven by secondary convective instabilities that are catalyzed by the primary
instability and thus act as the most significant injection scale for the ensuing
turbulent cascade and irreversible mixing, lies at the heart of the discussion to
follow. To make this clear, it is necessary to quantify the relative importance of
mixing events within the dynamically evolving flow as well as the evolution of
flow energetics, in particular, the exchanges between the kinetic energy and potential energy reservoirs. In Section 2, we therefore discuss a framework that may
be employed to achieve this, essentially by compartmentalization of the potential
energy of the flow into two parts, the background or minimum potential energy and
the available potential energy, as suggested by Winters and coworkers (Winters
et al. 1995, Winters & DAsaro 1996), whose methodology builds on the ideas
of Lorenz (1955). Furthermore, this framework is a generalization to three dimensions of a methodology suggested earlier by Thorpe (1977). It is important to
note that a somewhat similar approach was proposed independently by Scinocca
(1995).
Use of this compartmentalization, and appropriately defined mixing efficiencies that may be derived from it, has recently led to a clear understanding as to
why experimental measures of mixing efficiency have invariably led to estimates
on the order of 0.2. In Section 3 we summarize the results of numerical simulations of transitional shear flows, results that point toward the crucial importance
of late-time turbulence transition to understanding the mixing properties of stratified flows at sufficiently high Reynolds number and sufficiently low characteristic
Richardson number. In Section 4 we focus our attention on this late stage of flow
evolution in which transition occurs, demonstrating that mixing efficiency is expected to be a nonmonotonic function of the Richardson number, a result that
has extremely important and well-known dynamical consequences involving the
1 Nov 2002
18:38
140
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
(2a)
ui
= 0,
xi
(2b)
1 20
D 0
=
,
Dt
Re Pr x2j
(2c)
where 0 and p0 are the perturbation density and pressure, respectively, Pr is the
Prandtl number /, Re = V0d0/ is the Reynolds number, and Ri0 is the bulk
Richardson number g o d0/V20 , an overall measure of the relative importance of
the stratification compared to the inertia of the flow. With the appropriate boundary
conditions described above, the evolution of the kinetic energy K per unit volume
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
141
(3a)
= H ,
(3b)
K = hhh(u2 + v2 + w2 )/2ix iy iz ,
(3c)
with
where angle brackets denote averaging over the subscripted coordinate direction.
Similarly the potential energy P of the flow governed by Equations 2ac satisfies
the evolution equation
dP
= H + Dp ,
dt
(4a)
P = Rio hhhzix iy iz
(4b)
with
and
Dp =
(4c)
where Z is the nondimensional height of the domain, and (bottom) and (top)
are the nondimensional densities at the bottom and top boundaries, respectively.
Dp is the rate at which the potential energy of a statically stable density distribution would increase in the absence of macroscopic fluid motion (essentially
through conversion of internal energy into potential energy), and so for sufficiently
deep flow domains this plays no significant dynamical role in flow evolution. We
wish to distinguish this increase from mixing because we prefer to consider mixing as involving irreversible changes in fluid properties that are directly associated
with macroscopic fluid motions, rather than those changes that would occur in a
diffusive, macroscopically quiescent, system.
The crucial term for the quantification of mixing in these equations is, of course,
the heat flux, or buoyancy flux H, which quantifies the exchange between the kinetic and potential energy reservoirs. However, fluid motions can increase the
potential energy of the system by two qualitatively different processes, namely,
by either mixing or stirring. Any lifting (on average) of anomalously dense
fluid by the flow (hence involving negative values of H) will lead to an increase
of potential energy. However, this increase is only irreversible (and so is appropriately identified with mixing) if diffusion acts on sufficiently small scales so as to
change the density of the lifted fluid parcel, as well as its neighbors. Such mixing
is inevitably irreversible and acts over the small length scales characteristic of diffusion. Otherwise, the lifted parcels may later fall under gravity, reconverting the
1 Nov 2002
18:38
142
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
temporarily stored potential energy into kinetic energy, a process that is identified
as stirring (a larger-scale process that is nondiffusive).
It should be clear, therefore, that instantaneous values of the buoyancy flux H
can be very misleading as a proxy for mixing, particularly in flows with significant
time variability. Irreversible increases in potential energy owing to fluid motion can
only be distinguished from the buoyancy flux through averaging over sufficiently
long timescales so that the reversible effects of stirring completely cancel out.
Clearly, such averaging is not always possible in the laboratory or in the analysis
of oceanographic or meteorological data. However, the formalism of Winters et al.
(1995) does enable the time-dependent separation of mixing from stirring processes
because variations in the background potential energy can be identified directly
with mixing and the diffusion of the mean profile quantified by Dp (which these
authors have referred to as 8i).
We define the dimensional background potential energy P B(t) as
PB (t) = hgB (z, t) ziz .
(5)
In Equation 5, B(z, t) is the notional one-dimensional density distribution generated (by volume conserving adiabatic rearrangement of fluid elements) from the
actual density distribution (x, y, z, t) at the time t that is everywhere statically
stable. In general, the background potential energy is the minimum potential energy
that the flow can have, and this energy cannot be extracted from the potential
energy reservoir to drive (macroscopic) fluid motions, i.e., energy stored in P B
is inaccessible to stirring. Following Lorenz (1955), we then define an available
potential energy P A that is, at least in principle, available to the kinetic energy
reservoir for stirring as
PA = P PB ,
(6)
where P is the instantaneous total potential energy of the flow as defined in Equation 4b.
P A > 0 if there exists any lateral variations in density or statically unstable
regions. In Figure 2, we provide a schematic illustration of just such an adiabatic
rearrangement of fluid elements, showing that the generation of B can involve
changes in shape, but not the volume of the individual elements, and that the total
potential energy (defined by the mean density profile) is always bounded below
by the background potential energy.
In the absence of macroscopic fluid motion, P B will increase through the effect
of Dp, as defined in Equation 4c, as the mean profile diffuses. Crucially, P B can also
increase through irreversible mixing processes, and thus it is possible to partition
the buoyancy flux into two distinct terms and to write distinct evolution equations
for P A and P B as
dPA
= H M,
dt
(7a)
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
143
Figure 2 Schematic diagram of the adiabatic redistribution of fluid parcels (with densities
in the range +/ 31) to evaluate the background density distribution B associated with
the minimal background potential energy P B of a flow. Note that the fluid parcels change
shape but not volume. Also shown are the vertical distributions of the mean density and
the background density distribution B. [Adapted from figure 3 in Caulfield & Peltier (2000),
used with permission.]
dPB
= M + DP ,
dt
(7b)
1 Nov 2002
18:38
144
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
must be intensified to enhance the flux. However, this is only half the story: In
order to enhance transport, the area over which the flux enhancement occurs must
also increase, implying the need for neighboring density isosurfaces to become
complicated through the development of small scales of motion. A dense parcel,
feeling a buoyancy force restoring it to its neutral buoyancy height, will tend to
cause both these effects: Local gradients in density will intensify owing to the parcels motion (thus enhancing diffusive flux), and also, critically, the conversion of
potential to kinetic energy will tend to introduce small scales, thus enhancing transport. As we discuss below, such behavior can be directly identified and understood
in transitional shear flows but can also be found in forced stratified turbulence, provided that the flow exhibits regions of overturning (e.g., see Diamessis & Nomura
2002).
Indeed, a particular attraction of this formalism is that it is exceedingly straightforward to implement in a numerical simulation, as suggested by Winters et al.
(1995). At any particular time step, sorting algorithms can be used to order by
their density the n individual fluid elements located in the n grid volume elements
in the computational domain. These fluid parcels can then be notionally placed
into a purely one-dimensional array, created by deforming the n grid volume elements (while conserving volume) into n thin horizontal layers extending across
the whole flow domain. [See Caulfield & Peltier (2000) for a more detailed discussion, generalizing Winters et al. (1995), who proposed using the nondeformed
grid, thus allowing for horizontal density variations and hence generating only an
approximate upper bound to B.] This notional deformed grid, corresponding to
the identification of a unique height with each of the fluid elements arrayed within
a monotonic density field, creates the background density profile B(z) and hence
P B. As originally pointed out by Winters et al. (1995), and discussed further by
Tseng & Ferziger (2001), such sorting and reordering is intimately related to construction of the probability density function for the fluid density. Numerically, this
procedure can be performed at every time step, and so the time evolution of P B
can be followed, thus identifying the time dependence of the irreversible mixing
that is occurring.
It is of course important to quantify not only the absolute amount of mixing that
occurs, but also its efficiency, i.e., the proportion of the kinetic energy lost by the
fluid that leads to mixing, as opposed to that which is lost to viscous dissipation.
In the discussion to follow, we make use of a definition of instantaneous mixing
efficiency, say, E i, that we define as
Ei =
M
.
M+
(8a)
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
145
(8b)
where the time interval may be selected so as to isolate the manner in which
mixing efficiency changes over a series of appropriately selected epochs during
flow evolution. Such a definition eliminates the diffusive contribution DP, as this
is not associated with the kinetic energy, and hence the macroscopic fluid motion
of the system.
This elimination allows an identification of the cumulative mixing efficiency,
integrated over a sufficiently long time, with the flux Richardson number as was
defined by Linden (1979), namely, as the fraction of the change of the available
kinetic energy which appears as the potential energy of the stratification. This
is appropriate because it allows for the long-term cancellation of the influence of
those large-scale processes that we refer to as stirring. However, the flux Richardson
number Rf is often defined as
Rf =
H
.
H +
(9a)
H
u0 w0 dV/dz
(9b)
(see, e.g., Strang & Fernando 2001), these two definitions being equivalent if
the system is closed, i.e., the turbulent kinetic energy is statistically steady and
dominates the dissipation occurring within the flow.
The flux Richardson number is a parameter that lies at the heart of the parameterizations of mixing within turbulent closure schemes (see, e.g., Mellor & Yamada
1982). Here, we are interested in the mechanisms responsible for causing, and the
quantification of, stratified mixing within transitional stratified shear flows, and
in such fundamentally time-dependent circumstances, Rf would be an adequate
definition of mixing efficiency if the entire increase of potential energy caused by
the vertical flux of buoyancy H were irreversibly converted into an increase of
background potential energy P B. However, it is inevitable that some (often very
significant) fraction of H is simply associated with reversible stirring, and so it
should be clear that our definitions of mixing efficiency are most appropriate when
the required data are available, as is the case in numerical simulations.
Our procedure for the determination of mixing efficiency in stably stratified
parallel flows reviewed above is essentially a three-dimensional generalization of
the one-dimensional reordering proposed by Thorpe (1977) to assess the size of
overturning regions from one-dimensional observational profiles, generating the
so-called Thorpe scale LT. His scale is based on the reordering of individual density
measurements in such a way as to generate a stable vertical profile. Such sorting
enables one to generate an array of Thorpe displacements j for each of the
30 Nov 2002
9:43
146
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
density parcels. From these, one defines the Thorpe scale LT = ( 2j )1/2 as the rms
of the displacements. Flows in which mixing is intense are then flows in which
the Thorpe scale is large. Smyth & Moum (2000b) have examined in detail the
connection between the Thorpe scale and the characteristic scales of reordering
within a three-dimensional framework.
As described below, construction of the subdivision of the potential energy into
P B and P A, together with a consideration of the time-dependent evolution of these
fields, the quantities M and S = H M, and the mixing efficiencies defined
in Equation 8, yield valuable insight into the nature of the mixing that occurs in
stratified shear flows.
V(z)
= Vo tanh (z/d),
(10a)
(z)
= a + o tanh (Rz/d),
(10b)
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
147
in which R = 1.1 is the ratio of the thickness of the shear layer to the thickness of the
density inversion and is chosen to enable close accord with conditions achievable in
laboratory experiments (e.g., Thorpe 1987) and to be susceptible to instabilities of
KH type. This flow has been considered by many authors (Caulfield & Peltier 1994,
2000; Palmer et al. 1994, 1996; Cortesi et al. 1998; Smyth & Moum 2000a,b). We
briefly review what we perceive to be the key characteristics of the transition to a
stage of intense turbulent mixing in these flows, focusing on the critical importance
of the secondary convective instability to flow evolution as well as the variation of
the mixing characteristics with stratification. For these profiles, the initial gradient Richardson number is minimum at the midpoint of the shear layer and equals
Ri(0) = R Ri0, where Ri0 is the bulk Richardson number of the initially parallel
flow. We discuss eight different simulations in detail here, all with Pr = 1; Re =
750; and Ri(0) = 0, 0.025, 0.05, 0.0625, 0.075, 0.0875, 0.1, and 0.125. This
Reynolds number is comparable with that usually realized in laboratory experiments. Such a detailed investigation of the isolated effect of the variation in stratification enables us to identify clearly the nature of the nonmonotonicity of the
variation of the mixing efficiency with stratification, although it is important to
stress that it is also of great interest to understand the effect of varying the other
characteristic flow parameters (see Klaassen & Peltier 1985a; Smyth & Moum
2000a,b; Smyth et al. 2001 for a discussion of the effect of variation in Pr). We
return to this point in Section 5.
We have also chosen the density stratification to vanish in the region exterior
to the region of strong shear. When the exterior region is also stably stratified,
internal waves may be emitted from the region of turbulent collapse and the nature
of the evolution of the turbulent flow may be significantly affected. This interesting
source of complexity is not addressed here (see Sutherland & Peltier 1994, 1995;
Sutherland et al. 1994; Werne & Fritts 1999 for consideration of the behavior of
flows in which the influence of external stratification is included).
For all the flows that we consider, the primary KH instability develops in
a completely two-dimensional manner, with an initial exponential growth rate
that is well predicted by linear theory and that eventually saturates at nondimensional time t 4060, with this saturation time being dependent on initial
stratification. As discussed in detail by Caulfield & Peltier (2000), the roll-up
of the primary billow leads to significant increases in the available potential
energy P A, which is associated with the development of the statically unstable regions at the periphery of the primary billow (as is apparent in Figure 1).
There also is typically little irreversible mixing before the primary billow saturates. At this saturation time, the primary billows are usually very coherent,
although the outer-scale Reynolds number (defined by the vertical extent of the
billow and the total velocity jump) is on the order of 104 and so is sufficiently
high for us to expect a turbulence transition (see Dimotakis 2000 for further discussion). This is naturally due to the evolution of the flow in two dimensions:
As the flow becomes three dimensional, its character changes qualitatively (see
below).
1 Nov 2002
18:38
148
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
(11a)
(11b)
(12a)
3D (y, z, t) = 3D
(y, z)e t ,
(12b)
1
u 3D
u
+
+ w2D 3D = i p3D
D2 u 3D
y
z
Re
(13a)
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
149
v3D + v2D
p
1
v3D
v
v2D
v2D
+ w2D 3D + v3D
+ w3D
= 3D +
D2 v3D
y
z
y
z
y
Re
(13b)
w3D + v2D
+ v2D
3D
w3D
w
w2D
w2D
+ w2D 3D + v3D
+ w3D
=
y
z
y
z
1
p3D
+
Rio 3D
D2 w3D
z
Re
(13c)
1
3D
2D
2D
+ w2D 3D + v3D
+ w3D
=
D2 3D
y
z
y
z
Re Pr
i u3D +
v3D
w3D
+
= 0,
y
z
(13d)
(13e)
where the operator D2 = 2 + 2 / y 2 + 2 /z 2 . Because the pressure perturbation may be determined on the basis of the following diagnostic equation,
"
D2 p3D
= 2
v2D
y
w2D
+
y
v3D
y
v3D
z
w2D
z
#
Rio
w3D
z
v2D
z
w3D
y
3D
,
y
(13f)
we may eliminate p3D from the set (Equation 13). Thus, u3D is also removed,
thereby reducing the eigenvalue problem for to one involving only v3D , w3D , and
3D
. This system may be solved using a Galerkin method, which has the advantage
of rapid convergence.
To implement the Galerkin methodology, we expand the unknown fields on a
basis of orthogonal functions chosen such that the boundary conditions are satisfied
automatically. Specifically, we take
=
v3D , w3D , 3D
+ X
X
(14)
m= n=o
30 Nov 2002
9:44
150
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
mn
mn
Apq = hAAimn
pq Amn + hABipq Bmn + hACipq Cmn ,
(15a)
mn
mn
Bpq = hBAimn
pq Amn + hBBipq Bmn + hBCipq Cmn ,
(15b)
mn
mn
Cpq = hCAimn
pq Amn + hCBipq Bmn + hCCipq Cmn .
(15c)
and
Expressions for the four-dimensional coefficient arrays hAAi, hABi, etc., are provided by Smyth & Peltier (1994). By imposing a particular truncation scheme
upon the Galerkin expansions (Equation 14), specifically we use 2|m|+ n N,
and by taking N = 41 so that adequate resolution is achieved, Equation 15 then
poses a standard matrix eigenvalue problem for the eigenvalue . A single eigenvector is defined simply by concatenating the unknowns Apq, Bpq, and Cpq into a
single vector V with a single running subscript. Potylitsin & Peltier (1998, 1999)
have illustrated the eigenfunctions obtained in such analyses, including the impact
of background rotation parallel to the axis of the KH vortex tubes, by directly
contouring the three-dimensional perturbation vorticity field.
In Figure 4, we plot the real part of the eigenvalue , the growth rate r, as a
function of the spanwise wavenumber of the fastest growing three-dimensional
perturbation for Ri(0) = 0, 0.025, 0.05, and 0.1. On each diagram, the globally
most unstable mode is shown as the open circle. In the unstratified case, the growth
rate for the subdominant elliptical core mode is also indicated. Figure 5 shows the
cross-stream average over a wavelength 2/ of the perturbation kinetic energy
K3D determined on the basis of the eigenfunction of the most unstable mode from
K3D =
1 2
|u 3D | + |v3D |2 + |w3D |2 x .
2
(16)
Inspection of these results demonstrates that the fastest growing mode of threedimensional secondary instability becomes more sharply confined to the statically
unstable periphery of the individual billow cores as the strength of the ambient
stratification increases. Only in the limit where the background stratification vanishes entirely does the mode become focused on the hyperbolic stagnation points
that are located between adjacent billow cores. In Figure 5a, we show not only
K3D for the hyperbolic mode, but also K3D for the elliptical core mode by Pierrehumbert & Widnall (1982), the latter denoted by the heavy solid contours in the
billow core of the unstratified flow.
That these predictions of the linear stability analysis of finite amplitude KH
billows are precisely borne out by complete DNS solutions of the initial value
problem is demonstrated in Figure 6, where isosurfaces of spanwise vorticity
and streamwise vorticity from two typical three-dimensional DNS integrations
over the range of time during which the streamwise vortex streaks have just
emerged from the background two-dimensional KH flow of finite amplitude are
shown. Results for both the unstratified flow and the stratified flow for which
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
151
Figure 4 Predicted growth rate r of the most unstable three-dimensional mode against
spanwise wavenumber for Ri(0) = (a) 0, (b) 0.025, (c) 0.05, and (d ) 0.1. In each case, the
globally most unstable mode (and for the unstratified flow, the most unstable core-centered
mode) is marked with a circle.
the strength of the stratification is characterized by Ri(0) = 0.05 are also shown.
Intercomparison of the results for the unstratified flow (top) and the stratified flow
(bottom) demonstrates that the distinctly different regions of localization of the
secondary instability that is predicted by the Floquet analysis is precisely borne
out by the DNS results. In the unstratified case, streamwise vortex streaks originate through instability focused on the hyperbolic stagnation points that separate
adjacent billow cores, whereas in stratified flow, these streaks originate through
convective instability in the regions of overturned isopycnals. Their development
is triggered by a single convective overturning (releasing available potential energy
to the kinetic energy reservoir) in the statically unstable regions at the periphery of
the primary cores. Their growth is also driven by the background shear, with the
relative importance of these two processes depending on the overall stratification
(see Caulfield & Peltier 2000 for further discussion). Furthermore, the critical importance of these streamwise vortices for the transition to turbulence is confirmed
1 Nov 2002
18:38
152
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
Figure 5 Contours of K3D (as defined in Equation 16) for the most unstable threedimensional mode in flows with Ri(0) = (a) 0 (thin solid lines show the most unstable
(hyperbolic) mode with = 2.5, and thick solid lines show the most unstable elliptical mode
with = 0.9); (b) 0.025; (c) 0.05; and (d ) 0.1. [Adapted from figure 7 in Caulfield & Peltier
(2000), used with permission.]
x (x, y, z, t) =
+N
X
n=N
Cn (n , y, z, t)ei nx ,
(17a)
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
153
Figure 7 Variation of the various components of the normalized power spectrum density
P( n, t) (defined in Equation 18) against n at four characteristic times within a flow with
Ri(0) = 0.05. The wavenumber of the most unstable spanwise mode shown in Figure 4 is
denoted by a thick vertical dashed line.
with
Z
Cn (n , y, z, t) =
Lx
x ei nx dx
(17b)
(18)
1 Nov 2002
18:38
154
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
(19a)
K = h(V 2 )/2iz ,
(19b)
with
and
KKH =
v2KH + w2KH
2 y z,
(19c)
and the three-dimensional deviation from this cross-stream average, which we are
denoting by K3D and which is defined as
K3D = K K KKH .
(19d)
This compartmentalization allows us to isolate the evolution of three dimensionality from the evolving two-dimensional KH billow component of the flow.
Inspection of these diagnostics will demonstrate (from Figure 8a) that it is not
until nondimensional time t > 100 that total kinetic energy begins to decrease
rapidly owing to the intense viscous dissipation that onsets only with the proliferation of small spatial scales that develop during three-dimensional turbulent
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
155
Figure 8 Time evolution of (a) K(t)/K(0), (b) KKH(t)/K(0), and (c) K3D(t)/K(0) for threedimensional (as defined in Equations 3 and 19) simulations with Ri(0) = 0 (dotted line),
Ri(0) = 0.025 (solid line), Ri(0) = 0.05 (dashed line), and Ri(0) = 0.1 (dot-dashed line). The
characteristic times given in the table are marked with a circle [Ri(0) = 0], a plus [Ri(0) =
0.025], a cross [Ri(0) = 0.05], or an asterisk [Ri(0) = 0.1]. [Adapted from figure 8 in
Caulfield & Peltier (2000), used with permission.]
1 Nov 2002
18:38
156
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
Figure 9 Time variation of P (t)-DPt-P (0) (solid line) (as defined in Equation 4), P B(t)-DPtP B(0) (dashed line) (as defined in Equation 5), and P A(t) (dotted line) (as defined in Equation
6) for inherently three-dimensional simulations with Ri(0) = (a) 0.025, (b) 0.05, (c) 0.0625,
(d ) 0.075, (e) 0.0875, ( f ) 0.1, and (g) 0.125, compared with the time variation of E i (solid
line) and E c (dashed line) (as defined in Equations 8ab) for inherently three-dimensional
simulations with Ri(0) = (h) 0.025, (i) 0.05, ( j) 0.0625, (k) 0.075, (l ) 0.0875, (m) 0.1, and
(n) 0.125. The characteristic times shown in Figure 5 are marked with thick vertical dotted
lines.
stratified flows. The three key times discussed above in connection with the kinetic
energy diagnostics are shown on each frame as the vertical dotted lines.
Directly after the saturation of the primary billows, for each simulation there
is a phase of early time, preturbulent [using the nomenclature by Smyth et al.
(2001)] mixing that is extremely efficient. The instantaneous efficiencies, which
maximize subsequent to the time the two-dimensional waves saturate, reveal the
influence of the nutation and oscillation of the billow core that follows saturation.
Instantaneous mixing efficiency is observed to equal or exceed the high value of
0.75 near the time that the KH wave achieves maximum amplitude even though
the flows remain highly coherent. It is apparent that P B only starts to increase
once P A is large, and indeed much of the increase in P B occurs when the total
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
157
potential energy is decreasing (and so H > 0 as defined in Equation 3). Thus the
nave definition of the flux Richardson number given by Equation 9a is not useful.
At these Reynolds numbers, this early-stage mixing is associated with the convectively unstable regions around the periphery of the primary vortex core, which
naturally have the required characteristics for enhanced mixing (as discussed in
Section 2), i.e., regions of static instability associated with a source of P A, and
locally intensified density gradients. At sufficiently high Reynolds number, this region of the flow experiences enhanced small-scale mixing, as the available potential
energy is imperfectly converted back into kinetic energy. This mixing is primarily
due to the convective overturning that triggers the growth of the three-dimensional
perturbations, in the regions around the periphery of the primary billow core (see
Caulfield & Peltier 2000, where this identification could be made by the analysis
of spatially compartmentalized three-dimensional perturbation growth), and nucleates the characteristic streamwise vortices, which, at sufficiently high Reynolds
number, qualitatively change the long-time behavior of the flow.
Indeed, particularly for flows with weaker initial stratification, the total potential energy increases significantly at the same time as the total kinetic energy
begins to decrease significantly (compare with Figure 8). It is important to note
that this increase in potential energy at late time is not primarily due to a reversible
exchange between the kinetic energy and the available potential energy reservoirs.
Rather, the available potential energy P A exhibits strong variability (necessary to
enable irreversible mixing to occur) but of relatively small amplitude. The observed strong monotonic increase of total potential energy, at times later than the
second characteristic time at which the dissipation starts to increase markedly, is
clearly due to rapid increase in the background potential energy P B of the flow,
a further indication that it is in this latter stage of flow development in which
significant three-dimensional irreversible mixing is occurring. In fact, enhanced
three-dimensional irreversible mixing continues unabated during the entire posttransitional stage of flow development.
An important question to address clearly concerns the reason why this turbulence transition occurs. Once the convective instability develops, a large body of
numerical evidence (see Palmer et al. 1996; Cortesi et al. 1998; Werne & Fritts
1999; Caulfield & Peltier 2000; Smyth & Moum 2000a,b; Smyth et al. 2001) points
toward the inevitable development of disordered turbulent flow, with the streamwise vortices being driven to finite amplitude principally through vortex stretching
by the mean shear, with the primary billows once again playing a primarily catalytic role.
It is apparent that the transition to turbulence is triggered by the catastrophic
breakdown of these streamwise vortices. Owing to their location, the advection
and amplification through stretching of the streamwise vortices by the background
shear leads inevitably to the streamwise vortices propagating toward the braid
region, and there comes an instant when these vortices collide. Such a finite amplitude subcritical interaction occurred in all the simulations analyzed here and
appears to be the inevitable precursor to the onset of truly turbulent motions,
1 Nov 2002
18:38
158
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
reinforcing the concept that it is the characteristic scale of the streamwise vortices,
and not the significantly larger primary billow scale, that is the dominant injection
scale for the turbulent cascade. Indeed, owing to the unsteadiness of the flow, it
appears not to be appropriate to consider the development of tertiary instabilities
of the streamwise vortices to be the trigger for transition. Rather, subcritical, finite
amplitude vortex-vortex interactions seem to be the primary cause. Indeed, the
numerical evidence appears to point to the requirement not only of adequate resolution of individual streamwise vortices, but also of sufficient spanwise expanse
so that multiple streamwise vortices can develop and interact during transition in
order to capture properly the transition behavior.
However, it is important to stress that there is some evidence (Arendt et al. 1997,
Fritts et al. 1998) that the streamwise vortices may support a family of propagating
disturbances, referred to as Kelvin twist waves (see Arendt et al. 1997, Fritts et al.
1998 for a full discussion). These disturbances, which propagate along the vortex
tubes [and appear to arise naturally as a part of any initial disturbance on the
vortex (as demonstrated by Arendt et al. 1997)], can lead to fragmentation and
depletion of the vortex tubes, thus triggering transition to smaller and smaller
scales of motion within the flow. Arendt et al. (1997) and Fritts et al. (1998) also
found that vortex-vortex interactions with longer wavelengths [associated with the
cooperative Crow (1970) instability of neighboring elliptical vortices] could also
induce turbulence, and it is not yet clear which of these competing mechanisms,
if either, plays an active role in the transition process.
However, it is clear that the turbulence transition is intimately related to the
streamwise vortices induced by convective instability, as first described by Klaassen
& Peltier (1985b), and that they are the primary mechanism by which enhanced
dissipation is triggered within stratified shear flows. The issue as to whether this
turbulence transition mechanism is the dominant mechanism in all stratified shear
flows is highly complex, with the relative importance of the various stages of flow
evolution depending on both the flow parameters and the strength of the initial
forcing. For example, if the flow is forced so that merging events can occur before
the transition to turbulence, the mixing associated with the coherent, and essentially two-dimensional, merging of neighboring vortices can constitute a dominant
part of the total mixing experienced within the flow, as discussed by Smyth et al.
(2001).
Also, and as is apparent for the more strongly stratified simulation [with
Ri(0) = 0.125], the mixing associated with the preturbulent overturning in the
convective layers is very intense and constitutes a significant proportion of the
total mixing experienced by this flow. As is apparent in Figure 8b, the growth of
the primary KH billow in this flow is relatively weak, owing to the stabilizing
effect of stratification on this simulation. Also, as is apparent in Figure 8c, increasing stratification also reduces the growth rate of the developing three-dimensional
perturbations, and their saturation amplitude also reduces as the stratification increases. Such weakening growth of the primary KH instability for Ri(0) > 0.1
is commonly observed in laboratory experiments (see Koop & Browand 1979,
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
159
Linden 1979, Thorpe 1987, Fernando 1991). However, the crucial influence of
sufficiently high stratification actually appears to be the reduction in the saturated
amplitude of the three-dimensional perturbations, which, we believe, is indicative
of a qualitative change in the character of the flow.
Such behavior is also observed in numerical simulations of flows with constant
shear and density gradient, which are likely to have certain points of connection
with the turbulent flows considered here because the characteristic scales of motion
are becoming small compared to the mean shear and density gradient. As identified
by several authors (Holt et al. 1992, Jacobitz et al. 1997, Diamessis & Nomura
2002), the growth of turbulent kinetic energy is inhibited by buoyancy forces for
Ri > 0.1 in such flows. Our results are strongly suggestive of an analogous behavior,
with a relatively short period of turbulent disordered motion occurring for the flows
with higher Ri(0), whereas the less strongly stratified simulations exhibit extended
periods of turbulent mixing, leading to much greater total mixing than occurs in
the initial preturbulent phase. The relative importance of these two phases is likely
to also depend on the Reynolds number of the flow, as lower Reynolds number
will tend to reduce the significance of the turbulent phase at a given stratification
from the high Re behavior that shows a demarcation around Ri(0) = 0.1.
Nevertheless, in all the simulations we discuss here, it is clear that after transition occurs the absolute value of the background potential energy begins to grow.
This is particularly noticeable for flows with smaller initial stratification, where the
preturbulent convective overturning has achieved little mixing, yet the streamwise
vortices that are induced by this overturning grow rapidly and then trigger significantly increased mixing and strong dissipation. This behavior, as is apparent from
the spectra shown in Figure 7, occurs at substantially smaller scale than both the
primary KH instability and the injection scale of the secondary convective rolls,
thereby further reinforcing the analogy with the constant shear/constant density
gradient simulations.
Based on the use of Equations 8a and 8b, we show in Figure 9hn the evolution of the mixing efficiency, both instantaneous (solid lines) and cumulative
(dashed lines), for each of the seven stratified flow examples for which we have
performed detailed analysis. In this range of time during which three-dimensional
motions have proliferated and intense viscous dissipation is occurring, a state that
is turbulent in the conventional sense, the instantaneous mixing efficiency has
decreased substantially toward the experimentally determined value of approximately 0.20.3 that is characteristic of mixing associated with shear instabilities
(e.g., Linden 1979, Figure 4, where the flux Richardson number has been averaged
over a sufficiently long timescale for equality to be obtained between Rf and E c).
Particularly for flows with lower characteristic stratification, this phase of flow
evolution dominates the overall mixing character of the ultimate flow. In the next
section, we discuss this late stage in more detail, drawing connections between
the properties of this late stage with experimental and theoretical investigations
of mixing within stratified flows, paying particular attention to the implications of
the variation of mixing efficiency with stratification during this turbulent phase.
1 Nov 2002
18:38
160
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
4. TURBULENT-MIXING EFFICIENCY AS A
FUNCTION OF RICHARDSON NUMBER
Because the late, turbulent stage of flow evolution appears to be so important, it
is therefore useful to redefine the cumulative mixing efficiency as a function of
time t subsequent to the onset of true three-dimensional turbulent mixing. This
quantity, denoted Etc , is shown in Figure 10a for several flows that differ from one
another in terms of their initial Richardson number. Inspection of these results
shows that there exists clear evidence of nonmonotonicity of the mixing efficiency
as a function of the initial Richardson number. Also shown in Figure 10bd are
the mean fields of density, horizontal velocity, and B at the time t t3Dmax = 40
for the three-dimensional flows with Ri(0) = 0.025, 0.05, and 0.1, respectively,
showing clear evidence of the development of an intermediate layer of well-mixed
fluid when Ri(0) = 0.05.
Averaging over the interval for which results are shown in Figure 10a for the
cumulative mixing efficiency, we obtain the efficiency versus initial Richardson
numbered diagram shown in Figure 11a. Here we have plotted the results obtained
for seven nonzero values of the initial Richardson number. Inspection of the results
plotted on Figure 11a clearly shows that the efficiency with which the density
field is mixed in the turbulent regime is a strongly nonmonotonic function of the
stratification and moreover that the range of efficiencies that we have inferred is
in accord with the values that have been obtained from laboratory measurements.
As originally conjectured by Phillips (1972) and Posmentier (1977), nonmonotonicity in the mixing efficiency should lead to the development of distinct layers
within the stratified fluid. Physically, this can be understood (see Linden 1979 for
a clear discussion) by considering a location where the density gradient is locally
increased relative to neighboring values. If the mixing efficiency is a decreasing
function of stratification in this vicinity, the flux of buoyancy above and below this
particular level will be greater, where the stratification is weaker, thus leading to
accumulation at this location and hence intensification of the local gradient.
Such nonmonotonicity and dependence of the mixing efficiency purely on the
overall stratification requires that the dominant scales of mixing should be small
compared to the density gradient. Although this is not true of the early stages of
shear-flow development (where the characteristic scale of the flow is that of the
primary billow), the evidence from the numerical simulations points to the scales
of mixing at later times being substantially smaller than the injection scale of
the streamwise vortices characteristic of the secondary convective instability, thus
indeed being smaller than the stratification scales.
It is unfortunate that the simple physical model presented by Linden (1979)
leads to a mathematically ill-posed problem, as pointed out by Barenblatt et al.
(1993). They showed, however, that if a finite time lag existed for the turbulence to
adjust to the ambient stratification, then the problem became well posed, leading
to the development of finite layers. Consideration of the late-time behavior of
a stratified shear layer points to just such a finite time lag, as can be inferred
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
161
Figure 10 (a) Variation of the posttransition cumulative mixing efficiency (Ect ) with t-t3dmax
(listed in Figure 8) for inherently three-dimensional simulations with Ri(0) = 0.025 (solid
line), 0.05 (dashed line), and 0.1 (dot-dashed line). Vertical profiles of horizontally averaged
velocity (dotted line), horizontally averaged density (solid line), and background density B
(dot-dashed line) at the time t-t3dmax = 40 (listed in Figure 8) for three-dimensional flows
with Ri(0) = (b) 0.025, (c) 0.05, and (d) 0.1. [Adapted from figure 10 of Caulfield & Peltier
(2000), used with permission.]
from the mixing data shown in Figure 9. As already mentioned, nonzero available
potential energy is a necessary condition for irreversible mixing processes to occur
within the fluid, and such intermediate creation of P A occurs on the timescale
of the turnover of the turbulent eddies within the flow, yielding a mechanism
consistent with that postulated by Barenblatt et al. As is clearly demonstrated in
Figure 10, an intermediate well-mixed layer does indeed develop for intermediate
stratifications.
A further important point that needs to be noted is that there is a qualitative
change in the behavior of the flow for larger values of Ri(0), with the efficiency
increasing again at higher stratifications. This appears to be due to the fact that the
dissipation is decreasing owing to the reduced intensity of the streamwise vortices
at late times, whereas the quantity of mixing associated with this breakdown does
1 Nov 2002
18:38
162
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
Figure 11 (a) Variation of the time-averaged value of the posttransition cumulative mixing
efficiency (Ect ) with Richardson number for the seven simulations shown in Figure 9 (the
connecting curve is generated by cubic spline interpolation). (b) Schematic illustration of
three possible relations between the buoyancy gradient and buoyancy flux for turbulentmixing experiments as proposed by Balmforth et al. (1998). The three curves correspond to
the assumption that forcing exerts constant force per unit mass (dashed line), with constant
power (dotted line) or equipartition (solid line) implying turbulent eddies adjust to an
external velocity scale on the eddy turnover timescale. [Figure 11b is adapted from figure 1
in Balmforth et al. (1998), used with permission.]
not decrease, owing at least in part to the increased stratification and, hence, to the
increased characteristic density differences in the flow.
This behavior is somewhat suggestive of the variation of buoyancy flux with
overall stratification that was recently conjectured by Balmforth et al. (1998) in an
attempt to develop a model to explain the layer formation observed in experiments
attributed to Park et al. (1994). Balmforth et al. (1998) postulated, using mixinglength theory, that such a variation in buoyancy flux (as shown in Figure 11b) could
occur if the characteristic eddy speed for turbulent motions within a stratified fluid
adjusted to an external driving velocity (in this case, that of the stirring rods)
on the timescale of an eddy turnover. Within the framework of a stratified shear
flow, the analogous assumption is that the streamwise vortices, associated with
the convective instability, are intensified (and hence driven to mixing) by the
background shear on a timescale comparable with their turnover time, essentially
that which occurs during the relatively rapid period of turbulence onset within the
flow. This gives further support to the observation that stratified mixing driven
by the breakdown of secondary streamwise vortices may lead to the creation of a
staircase density structure.
5. CONCLUSIONS
In this paper, we review and somewhat extend our own recent work on the inference
of mixing efficiency in stratified parallel flows that undergo the process of turbulent
collapse. This process begins with the onset of a two-dimensional, essentially
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
163
1 Nov 2002
18:38
164
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
165
LITERATURE CITED
Arendt S, Fritts DC, Andreassen O. 1997.
The initial value problem for Kelvin vortex
waves. J. Fluid Mech. 344:181212
Balmforth NJ, Llewellyn Smith SG, Young
WR. 1998. Dynamics of interfaces and layers
in a stratified turbulent fluid. J. Fluid Mech.
355:32958
Barenblatt GI, Bertsch M, Dal Paso R, Prostokishin VM, Ughi M. 1993. A mathematical
model of turbulent heat and mass transfer in
stably stratified shear flow. J. Fluid Mech.
253:34158
Browand FK, Winant CD. 1973. Laboratory observations of shear instability in a stratified
fluid. Bound.-Layer Meteorol. 5:6777
Caulfield CP, Kerswell RR. 2001. Maximal
mixing rate in turbulent stably stratified Couette flow. Phys. Fluids 13:894900
Caulfield CP, Peltier WR. 1994. Threedimensionalization of the stratified mixing
layer. Phys. Fluids 6:38035
Caulfield CP, Peltier WR. 2000. The anatomy
of the mixing transition in homogeneous and
stratified free shear layers. J. Fluid Mech.
413:147
Cortesi AB, Yadigaroglu G, Bannerjee S. 1998.
Numerical investigation of the formation of
1 Nov 2002
18:38
166
AR
AR159-FM35-08.tex
PELTIER
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
CAULFIELD
wave. Part 2. Vortex interactions and transition to turbulence. J. Fluid Mech. 367:4765
Gerz T, Schumann U, Elghobashi SE. 1989. Direct numerical simulation of stratified homogeneous turbulent shear flows. J. Fluid Mech.
200:56394
Holmboe J. 1962. On the behaviour of symmetric waves in stratified shear layers. Geofys.
Publ. Oslo 24:67113
Holt SE, Koseff JR, Ferziger JH. 1992. A numerical study of the evolution and structure
of homogeneous stably stratified sheared turbulence. J. Fluid Mech. 237:499539
Howard LN. 1961. Note on a paper of John W.
Miles. J. Fluid Mech. 10:50912
Jacobitz FG, Sarkar S, Van Atta CW. 1997. Direct numerical simulations of the turbulence
evolution in a uniformly sheared and stably
stratified flow. J. Fluid Mech. 342:23161
Keller KH, Van Atta CW. 2000. An experimental investigation of the vertical temperature
structure of homogeneous stratified shear turbulence. J. Fluid Mech. 425:129
Kerswell RR. 2002. Elliptical instability. Annu.
Rev. Fluid Mech. 34:83113
Klaassen GP, Peltier WR. 1985a. Prandtl number effects on the evolution and stability of
finite amplitude Kelvin-Helmholtz billows.
Geophys. Astrophys. Fluid Dyn. 32:2360
Klaassen GP, Peltier WR. 1985b. The onset
of turbulence in finite amplitude KelvinHelmholtz billows. J. Fluid Mech. 155:135
Klaassen GP, Peltier WR. 1989. The role of
transverse secondary instabilities in the evolution of free shear layers. J. Fluid Mech.
202:367402
Klaassen GP, Peltier WR. 1991. The influence
of stratification on secondary instabilities in
free shear layers. J. Fluid Mech. 227:71106
Koop G, Browand FK. 1979. Instability and turbulence in stratified fluids. J. Fluid Mech.
93:13559
Lawrence GA, Browand FK, Redekopp LG.
1991. The stability of a sheared density interface. Phys. Fluids A 3:236070
Linden PF. 1979. Mixing in stratified fluids.
Geophys. Astrophys. Fluid Dyn. 13:223
Lorenz EN. 1955. Available potential energy
1 Nov 2002
18:38
AR
AR159-FM35-08.tex
AR159-FM35-08.sgm
LaTeX2e(2002/01/18)
P1: IBC
167
29 Nov 2002
13:41
AR
AR159-08-COLOR.tex
AR159-08-COLOR.SGM
LaTeX2e(2002/01/18)
P1: GDL
13:41
AR
AR159-08-COLOR.tex
AR159-08-COLOR.SGM
LaTeX2e(2002/01/18)
P1: GDL
29 Nov 2002
Figure 6 Isosurfaces of spanwise vorticity (green) and positive (blue) and negative
(red) streamwise vorticity within (a) unstratified and (b) stratified (with Ri(0) = 0.05)
simulations containing two neighboring primary Kelvin-Helmholtz billows, during the
development of intense spanwise-periodic streamwise vortices.
P1: FDS
11:16
Annual Reviews
AR159-FM
CONTENTS
STANLEY CORRSIN: 19201986, John L. Lumley and Stephen H. Davis
AIRCRAFT ICING, Tuncer Cebeci and Fassi Kafyeke
WATER-WAVE IMPACT ON WALLS, D. H. Peregrine
MECHANISMS ON TRANSVERSE MOTIONS IN TURBULENT WALL
FLOWS, G. E. Karniadakis and Kwing-So Choi
INSTABILITIES IN FLUIDIZED BEDS, Sankaran Sundaresan
AERODYNAMICS OF SMALL VEHICLES, Thomas J. Mueller
and James D. DeLaurier
1
11
23
45
63
89
113
135
169
183
229
267
295
317
341
373
413
441
469
vii
P1: FDS
viii
11:16
Annual Reviews
AR159-FM
CONTENTS
INDEXES
Subject Index
Cumulative Index of Contributing Authors, Volumes 135
Cumulative Index of Chapter Titles, Volumes 135
ERRATA
497
521
528